A Temporal Boosted YOLO-Based Model for Birds Detection around Wind Farms

Abstract

Object detection for sky surveillance is a challenging problem due to having small objects in a large volume and a constantly changing background which requires high resolution frames. For example, detecting flying birds in wind farms to prevent their collision with the wind turbines. This paper proposes a YOLOv4-based ensemble model for bird detection in grayscale videos captured around wind turbines in wind farms. In order to tackle this problem, we introduce two datasets—(1) Klim and (2) Skagen—collected at two locations in Denmark. We use Klim training set to train three increasingly capable YOLOv4 based models. Model 1 uses YOLOv4 trained on the Klim dataset, Model 2 introduces tiling to improve small bird detection, and the last model uses tiling and temporal stacking and achieves the best mAP values on both Klim and Skagen datasets. We used this model to set up an ensemble detector, which further improves mAP values on both datasets. The three models achieve testing mAP values of 82%, 88%, and 90% on the Klim dataset. mAP values for Model 1 and Model 3 on the Skagen dataset are 60% and 92%. Improving object detection accuracy could mitigate birds’ mortality rate by choosing the locations for such establishment and the turbines location. It can also be used to improve the collision avoidance systems used in wind energy facilities.

Automatic Detection of Feral Pigeons in Urban Environments Using Deep Learning

“…Secondly, most existing studies on bird detection using deep learning techniques lack a specific focus on individual bird species, such as feral pigeons. Some studies concentrate on the accurate identification of various bird species in airborne scenarios [ 14 , 15 , 24 , 46 ], while others explore the classification and detection of different bird species in natural environments, such as wind farms or aquatic habitats [ 19 , 45 ]. Additionally, a few studies specifically investigate the detection and counting of different bird species in specific regions, such as birds on transmission lines [ 18 ].…”

“…Through a series of experiments, we demonstrate the effectiveness of our proposed method in feral pigeon detection in urban areas. In bird detection, most studies have utilized one-stage (YOLO) [ 15 , 17 , 19 , 21 ] and two-stage (Faster R-CNN and Mask R-CNN) [ 43 , 44 ] object detection models. The original Mask R-CNN has demonstrated great performance in bird detection [ 37 ].…”

“…Therefore, we replace the backbone of the original Mask R-CNN with the Swin Transformer and add FPN as the network’s neck for multi-scale feature fusion [ 51 ]. After adjusting the network, to evaluate the performance of the Swin-Mask R-CNN model, we compare it with commonly used object detection methods for bird detection, including YOLO [ 16 , 18 , 19 , 21 ], Faster R-CNN [ 45 ], Mask R-CNN [ 46 ], and our proposed method (Swin-Mask R-CNN) on our feral pigeon dataset. The mAP of our proposed Swin-Mask R-CNN model reaches the highest value of 0.68.…”

“…A model dubbed DC-YOLO, based on YOLOv3, was devised to accurately detect bird populations near power lines [ 17 ], and one modified YOLO model was used to detect cage-free hens on the litter floor [ 18 ]. Furthermore, a temporal boosted YOLO model was constructed for detecting birds in specific wind farms [ 19 ], and a combination of YOLO and Kalman filter [ 20 ] was applied to low-light images for detecting and tracking chickens [ 21 ]. These methods have improved bird detection accuracy.…”

Research on the feasibility of object detection technology in airport airspace control

“…This can be done using object detection techniques in artificial intelligence. The technology already has precedents in other areas, such as the use of object detection technology around wind farms to prevent bird strikes [1].…”